cs.ROAug 28, 2026

PanelShield: Verifiable Closed-Loop Safe Planning for Robotic Industrial Panel Operation

Authors: Guipeng Xin, Jiahe Xu, Chenhui Wan, Jie Liu, Youmin Hu, Zhongxu Hu

Organizations: School of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan, China

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Industrial panel operation is knowledge-intensive and safety-critical. Beyond control recognition and action generation, execution must satisfy constraints in operation manuals and safety regulations. While foundation-model-based planners show strong semantic capability, they typically lack computable, localizable, and reproducible mechanisms for violation detection and repair. To address this, we propose PanelShield, a verifiable closed-loop safety planning framework for manual-guided industrial panel operation. The framework generates parameterized action primitive sequences from task-relevant manual evidence and applies dual formal verification with LTL and a Safety FSM to enforce cross-step temporal correctness and local transition legality. When violations occur, it outputs a structured counterexample with the earliest violating step and cause, enabling targeted repair and re-verification. We build a multi-level long-horizon planning benchmark covering three representative industrial device panels, and evaluate the framework in simulation and real-world robotic experiments. Results show that PanelShield improves complex safety-constrained task performance over foundation-model-only planning baselines while reducing the violation rate to 2.7%, with 4.1 s total latency. Real-world experiments demonstrate end-toend feasibility. Overall, PanelShield offers a verifiable approach to robotic panel operation that balances flexibility, safety, and auditability.

Figures & tables

Fig. 1: Core challenge of industrial panel operation. Dispersed procedures and safety constraints jointly form a procedural constraint barrier that is difficult to handle directly. Traditional script-based methods are safe but rigid, while foundation-model-based planning is flexible but prone to violations. Our goal is to preserve flexibility while enabling verifiable safety planning with closed-loop error correction.
Fig. 2: PanelShield core loop for verifiable closed-loop planning in industrial panel operation. A VLM-based planner generates a candidate plan from instruction, device state, and task-relevant manual evidence; the plan is then converted into a computable run and checked by complementary LTL-based temporal verification and Safety-FSM-based legality verification. Violations trigger structured counterexample generation and feedback-guided repair, and the loop repeats until a final safe executable plan is obtained.
Fig. 3: Example of counterexample-guided closed-loop repair in PanelShield.
Fig. 4: Simulation planning baseline results on industrial panel operation tasks.
MethodRepair ↓\downarrowTotal Latency ↓\downarrowPlanning Time ↓\downarrowVerification Time ↓\downarrow
(avg. iters)(s)(s)(s)
LLM-only-1.71.7-
Prompt-Safety-2.22.2-
Post-hoc Judge3.94.53.41.1
Ours (PanelShield)3.64.13.50.6
TABLE I: Repair iterations and planning time overhead results.
MethodEarliest-violation Step ↓\downarrowLTL Violations ↓\downarrowFSM Violations ↓\downarrowAvg Edit Size / Repair ↓\downarrow
(steps)(%)(%)(steps)
Full (LTL + FSM + Repair)2.12.00.71.3
w/o LTL (FSM + Repair)2.4-1.61.8
w/o FSM (LTL + Repair)2.52.8-1.9
w/o Repair (LTL + FSM, no repair)2.02.10.8-
Post-hoc Judge3.62.41.13.0
TABLE II: Ablation results on violation localization, violation decomposition, and repair locality.
Fig. 5: Real-world chained-task execution example.
TaskInference Time (s) ↓\downarrowExecution Time (s) ↓\downarrow
1. Turn on main power3.55.8
2. Switch low speed under PANEL4.25.8
3. Change the direction to REV3.86.1
4. Perform a jog operation3.74.2
5. Verify READY, then START5.610.7
TABLE III: Real-world results on long-horizon chained tasks.

Explore similar work

CardsList

  1. MaCoPlanner: LLM-Assisted Manual-Compiled Task Planning with Proactive Safety Verification for Robotic Industrial Panel Operation

    Aug 28, 2026Guipeng Xin, Jiahe Xua, Mohammad Deghat +4Robot Task PlanningLarge Language Model-Based Robot Planning

    School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan, China · School of Mechanical and Manufacturing Engineering, University of New South Wales, Sydney, Australia · School of Civil and Hydraulic Engineering, Huazhong University of Science and Technology, Wuhan, China

  2. SafeGen-LLM: Enhancing Safety Generalization in Task Planning for Robotic Systems

    Feb 27, 2026Jialiang Fan, Weizhe Xu, Mengyu Liu +3Large Language Model-Based Robot PlanningRobot Task Planning

    University of Notre Dame · Washington State University · University of Pennsylvania

  3. MANIGUARD: A Benchmark and Data Suite for Specification-Grounded Safety Evaluation and Improvement of Robotic Manipulation

    Aug 18, 2026Yiyan Peng, Philip Wang, Simon Sinong Zhan +11Robotic ManipulationSafety Evaluation of VLA Models

    Northwestern University · Stanford University · William & Mary